voixful-canary-qwen-2.5b

Apple Core AI (Palette4-quantized) conversion of nvidia/canary-qwen-2.5b — a SALM (Conformer + Qwen3) with the best punctuation/casing, for on-device speech-to-text on Apple Silicon. Used by the Voixful engine and PrivoVoice.

License & attribution

Weights derive from nvidia/canary-qwen-2.5b (CC-BY-4.0, © NVIDIA), redistributed under CC-BY-4.0 with attribution. The Voixful conversion pipeline is Apache-2.0.